MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has launched Muse Wonder and Muse Wonder Full, its most advanced artificial intelligence models so far. These systems are designed to assist with coding, managing extended AI agents, and handling complex personal or work-related tasks. Meta states that the models can support users in processing insurance claims, understanding leases, analyzing finances, and developing functional applications. Muse Wonder became accessible on October 12 via the Meta Model API and Muse Code, while Muse Wonder Full was introduced for Muse Code subscribers.
This launch signals Meta’s increased focus on AI systems capable of multi-step tasks, moving beyond simple prompt responses. Muse Wonder can plan lengthy projects, assign parts of the work to sub-agents, leverage software tools, and recover from errors during execution. The models are built for advanced computing across browsers, desktop programs, and mobile apps. Both support a context window of about one million tokens, enabling them to handle large data in a single session.
Meta incorporated two features named Active Memory and Active Focus to enable more extended and personalized interactions. Active Memory maintains relevant details across conversations and retrieves information when needed for new tasks. Active Focus allows the models to continue working despite difficulties, test alternative approaches, and recover from mistakes. The models can also seek clarification from users and verify actions before execution, especially for important tasks.
Models are designed for coding and persistent AI agents
Muse Wonder costs $4 per million input tokens and $20 per million output tokens via the Meta Model API. Cached input is priced at $0.20 per million tokens, according to the developer information. Muse Wonder Full is only available through paid Muse Code subscriptions, not separate API pricing. Developers with higher workloads can opt for Power Usage at $50 a month and Maximum Usage at $200 a month.
Meta reports that Muse Wonder Full outperforms many benchmarks related to agentic and coding tasks. Tests like DeepSWE and SWE-Marathon, which evaluate software engineering and long-term coding abilities, highlight its performance. Independent reviewer Artificial Analysis assigned Muse Wonder an Intelligence Index score of 52, and Muse Wonder Full scored 56 at maximum reasoning effort. These results place both models above Meta’s earlier Muse Spark generation on the index.
Enhanced safety features accompany expanded model capabilities
Meta trained the Muse Wonder series to prevent sharing private data without permission and to decline harmful requests. The models also include cybersecurity tools to help developers identify vulnerabilities in their code and improve defenses. Meta states that its most capable models are evaluated under the Meta Superintelligence Scaling Framework. The company plans to publish a more detailed report covering additional safety assessments after the initial launch.
Meta began a comprehensive overhaul of its AI efforts approximately 14 months before releasing Muse Wonder. The new models build on the foundation established by Muse Spark, with training conducted at its Prometheus data center. The company intends to integrate Muse Wonder into more features within its Muse platform, starting with subscribers. The October launch provides immediate access for developers, reserving the larger Muse Wonder Full for Muse Code members.